fs/exec.c

   1 /*
   2  *  linux/fs/exec.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * #!-checking implemented by tytso.
   9  */
  10 /*
  11  * Demand-loading implemented 01.12.91 - no need to read anything but
  12  * the header into memory. The inode of the executable is put into
  13  * "current->executable", and page faults do the actual loading. Clean.
  14  *
  15  * Once more I can proudly say that linux stood up to being changed: it
  16  * was less than 2 hours work to get demand-loading completely implemented.
  17  *
  18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19  * current->executable is only used by the procfs.  This allows a dispatch
  20  * table to check for several different types  of binary formats.  We keep
  21  * trying until we recognize the file or we run out of supported binary
  22  * formats.
  23  */
  24
  25 #include <linux/slab.h>
  26 #include <linux/file.h>
  27 #include <linux/fdtable.h>
  28 #include <linux/freezer.h>
  29 #include <linux/mm.h>
  30 #include <linux/stat.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/swap.h>
  33 #include <linux/string.h>
  34 #include <linux/init.h>
  35 #include <linux/pagemap.h>
  36 #include <linux/perf_event.h>
  37 #include <linux/highmem.h>
  38 #include <linux/spinlock.h>
  39 #include <linux/key.h>
  40 #include <linux/personality.h>
  41 #include <linux/binfmts.h>
  42 #include <linux/utsname.h>
  43 #include <linux/pid_namespace.h>
  44 #include <linux/module.h>
  45 #include <linux/namei.h>
  46 #include <linux/mount.h>
  47 #include <linux/security.h>
  48 #include <linux/syscalls.h>
  49 #include <linux/tsacct_kern.h>
  50 #include <linux/cn_proc.h>
  51 #include <linux/audit.h>
  52 #include <linux/tracehook.h>
  53 #include <linux/kmod.h>
  54 #include <linux/fsnotify.h>
  55 #include <linux/fs_struct.h>
  56 #include <linux/pipe_fs_i.h>
  57 #include <linux/oom.h>
  58 #include <linux/compat.h>
  59 #include <linux/sched.h>
  60 #include <linux/fs.h>
  61 #include <linux/path.h>
  62
  63 #include <asm/uaccess.h>
  64 #include <asm/mmu_context.h>
  65 #include <asm/tlb.h>
  66 #include "internal.h"
  67
  68 int core_uses_pid;
  69 char core_pattern[CORENAME_MAX_SIZE] = "core";
  70 unsigned int core_pipe_limit;
  71 int suid_dumpable = 0;
  72
  73 struct core_name {
  74         char *corename;
  75         int used, size;
  76 };
  77 static atomic_t call_count = ATOMIC_INIT(1);
  78
  79 /* The maximal length of core_pattern is also specified in sysctl.c */
  80
  81 static LIST_HEAD(formats);
  82 static DEFINE_RWLOCK(binfmt_lock);
  83
  84 int __register_binfmt(struct linux_binfmt * fmt, int insert)
  85 {
  86         if (!fmt)
  87                 return -EINVAL;
  88         write_lock(&binfmt_lock);
  89         insert ? list_add(&fmt->lh, &formats) :
  90                  list_add_tail(&fmt->lh, &formats);
  91         write_unlock(&binfmt_lock);
  92         return 0;
  93 }
  94
  95 EXPORT_SYMBOL(__register_binfmt);
  96
  97 void unregister_binfmt(struct linux_binfmt * fmt)
  98 {
  99         write_lock(&binfmt_lock);
 100         list_del(&fmt->lh);
 101         write_unlock(&binfmt_lock);
 102 }
 103
 104 EXPORT_SYMBOL(unregister_binfmt);
 105
 106 static inline void put_binfmt(struct linux_binfmt * fmt)
 107 {
 108         module_put(fmt->module);
 109 }
 110
 111 /*
 112  * Note that a shared library must be both readable and executable due to
 113  * security reasons.
 114  *
 115  * Also note that we take the address to load from from the file itself.
 116  */
 117 SYSCALL_DEFINE1(uselib, const char __user *, library)
 118 {
 119         struct file *file;
 120         char *tmp = getname(library);
 121         int error = PTR_ERR(tmp);
 122         static const struct open_flags uselib_flags = {
 123                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 124                 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
 125                 .intent = LOOKUP_OPEN
 126         };
 127
 128         if (IS_ERR(tmp))
 129                 goto out;
 130
 131         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
 132         putname(tmp);
 133         error = PTR_ERR(file);
 134         if (IS_ERR(file))
 135                 goto out;
 136
 137         error = -EINVAL;
 138         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 139                 goto exit;
 140
 141         error = -EACCES;
 142         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 143                 goto exit;
 144
 145         fsnotify_open(file);
 146
 147         error = -ENOEXEC;
 148         if(file->f_op) {
 149                 struct linux_binfmt * fmt;
 150
 151                 read_lock(&binfmt_lock);
 152                 list_for_each_entry(fmt, &formats, lh) {
 153                         if (!fmt->load_shlib)
 154                                 continue;
 155                         if (!try_module_get(fmt->module))
 156                                 continue;
 157                         read_unlock(&binfmt_lock);
 158                         error = fmt->load_shlib(file);
 159                         read_lock(&binfmt_lock);
 160                         put_binfmt(fmt);
 161                         if (error != -ENOEXEC)
 162                                 break;
 163                 }
 164                 read_unlock(&binfmt_lock);
 165         }
 166 exit:
 167         fput(file);
 168 out:
 169         return error;
 170 }
 171
 172 #ifdef CONFIG_MMU
 173 /*
 174  * The nascent bprm->mm is not visible until exec_mmap() but it can
 175  * use a lot of memory, account these pages in current->mm temporary
 176  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 177  * change the counter back via acct_arg_size(0).
 178  */
 179 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 180 {
 181         struct mm_struct *mm = current->mm;
 182         long diff = (long)(pages - bprm->vma_pages);
 183
 184         if (!mm || !diff)
 185                 return;
 186
 187         bprm->vma_pages = pages;
 188         add_mm_counter(mm, MM_ANONPAGES, diff);
 189 }
 190
 191 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 192                 int write)
 193 {
 194         struct page *page;
 195         int ret;
 196
 197 #ifdef CONFIG_STACK_GROWSUP
 198         if (write) {
 199                 ret = expand_downwards(bprm->vma, pos);
 200                 if (ret < 0)
 201                         return NULL;
 202         }
 203 #endif
 204         ret = get_user_pages(current, bprm->mm, pos,
 205                         1, write, 1, &page, NULL);
 206         if (ret <= 0)
 207                 return NULL;
 208
 209         if (write) {
 210                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 211                 struct rlimit *rlim;
 212
 213                 acct_arg_size(bprm, size / PAGE_SIZE);
 214
 215                 /*
 216                  * We've historically supported up to 32 pages (ARG_MAX)
 217                  * of argument strings even with small stacks
 218                  */
 219                 if (size <= ARG_MAX)
 220                         return page;
 221
 222                 /*
 223                  * Limit to 1/4-th the stack size for the argv+env strings.
 224                  * This ensures that:
 225                  *  - the remaining binfmt code will not run out of stack space,
 226                  *  - the program will have a reasonable amount of stack left
 227                  *    to work from.
 228                  */
 229                 rlim = current->signal->rlim;
 230                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 231                         put_page(page);
 232                         return NULL;
 233                 }
 234         }
 235
 236         return page;
 237 }
 238
 239 static void put_arg_page(struct page *page)
 240 {
 241         put_page(page);
 242 }
 243
 244 static void free_arg_page(struct linux_binprm *bprm, int i)
 245 {
 246 }
 247
 248 static void free_arg_pages(struct linux_binprm *bprm)
 249 {
 250 }
 251
 252 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 253                 struct page *page)
 254 {
 255         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 256 }
 257
 258 static int __bprm_mm_init(struct linux_binprm *bprm)
 259 {
 260         int err;
 261         struct vm_area_struct *vma = NULL;
 262         struct mm_struct *mm = bprm->mm;
 263
 264         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 265         if (!vma)
 266                 return -ENOMEM;
 267
 268         down_write(&mm->mmap_sem);
 269         vma->vm_mm = mm;
 270
 271         /*
 272          * Place the stack at the largest stack address the architecture
 273          * supports. Later, we'll move this to an appropriate place. We don't
 274          * use STACK_TOP because that can depend on attributes which aren't
 275          * configured yet.
 276          */
 277         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 278         vma->vm_end = STACK_TOP_MAX;
 279         vma->vm_start = vma->vm_end - PAGE_SIZE;
 280         vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 281         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 282         INIT_LIST_HEAD(&vma->anon_vma_chain);
 283
 284         err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
 285         if (err)
 286                 goto err;
 287
 288         err = insert_vm_struct(mm, vma);
 289         if (err)
 290                 goto err;
 291
 292         mm->stack_vm = mm->total_vm = 1;
 293         up_write(&mm->mmap_sem);
 294         bprm->p = vma->vm_end - sizeof(void *);
 295         return 0;
 296 err:
 297         up_write(&mm->mmap_sem);
 298         bprm->vma = NULL;
 299         kmem_cache_free(vm_area_cachep, vma);
 300         return err;
 301 }
 302
 303 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 304 {
 305         return len <= MAX_ARG_STRLEN;
 306 }
 307
 308 #else
 309
 310 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 311 {
 312 }
 313
 314 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 315                 int write)
 316 {
 317         struct page *page;
 318
 319         page = bprm->page[pos / PAGE_SIZE];
 320         if (!page && write) {
 321                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 322                 if (!page)
 323                         return NULL;
 324                 bprm->page[pos / PAGE_SIZE] = page;
 325         }
 326
 327         return page;
 328 }
 329
 330 static void put_arg_page(struct page *page)
 331 {
 332 }
 333
 334 static void free_arg_page(struct linux_binprm *bprm, int i)
 335 {
 336         if (bprm->page[i]) {
 337                 __free_page(bprm->page[i]);
 338                 bprm->page[i] = NULL;
 339         }
 340 }
 341
 342 static void free_arg_pages(struct linux_binprm *bprm)
 343 {
 344         int i;
 345
 346         for (i = 0; i < MAX_ARG_PAGES; i++)
 347                 free_arg_page(bprm, i);
 348 }
 349
 350 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 351                 struct page *page)
 352 {
 353 }
 354
 355 static int __bprm_mm_init(struct linux_binprm *bprm)
 356 {
 357         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 358         return 0;
 359 }
 360
 361 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 362 {
 363         return len <= bprm->p;
 364 }
 365
 366 #endif /* CONFIG_MMU */
 367
 368 /*
 369  * Create a new mm_struct and populate it with a temporary stack
 370  * vm_area_struct.  We don't have enough context at this point to set the stack
 371  * flags, permissions, and offset, so we use temporary values.  We'll update
 372  * them later in setup_arg_pages().
 373  */
 374 int bprm_mm_init(struct linux_binprm *bprm)
 375 {
 376         int err;
 377         struct mm_struct *mm = NULL;
 378
 379         bprm->mm = mm = mm_alloc();
 380         err = -ENOMEM;
 381         if (!mm)
 382                 goto err;
 383
 384         err = init_new_context(current, mm);
 385         if (err)
 386                 goto err;
 387
 388         err = __bprm_mm_init(bprm);
 389         if (err)
 390                 goto err;
 391
 392         return 0;
 393
 394 err:
 395         if (mm) {
 396                 bprm->mm = NULL;
 397                 mmdrop(mm);
 398         }
 399
 400         return err;
 401 }
 402
 403 struct user_arg_ptr {
 404 #ifdef CONFIG_COMPAT
 405         bool is_compat;
 406 #endif
 407         union {
 408                 const char __user *const __user *native;
 409 #ifdef CONFIG_COMPAT
 410                 compat_uptr_t __user *compat;
 411 #endif
 412         } ptr;
 413 };
 414
 415 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 416 {
 417         const char __user *native;
 418
 419 #ifdef CONFIG_COMPAT
 420         if (unlikely(argv.is_compat)) {
 421                 compat_uptr_t compat;
 422
 423                 if (get_user(compat, argv.ptr.compat + nr))
 424                         return ERR_PTR(-EFAULT);
 425
 426                 return compat_ptr(compat);
 427         }
 428 #endif
 429
 430         if (get_user(native, argv.ptr.native + nr))
 431                 return ERR_PTR(-EFAULT);
 432
 433         return native;
 434 }
 435
 436 /*
 437  * count() counts the number of strings in array ARGV.
 438  */
 439 static int count(struct user_arg_ptr argv, int max)
 440 {
 441         int i = 0;
 442
 443         if (argv.ptr.native != NULL) {
 444                 for (;;) {
 445                         const char __user *p = get_user_arg_ptr(argv, i);
 446
 447                         if (!p)
 448                                 break;
 449
 450                         if (IS_ERR(p))
 451                                 return -EFAULT;
 452
 453                         if (i++ >= max)
 454                                 return -E2BIG;
 455
 456                         if (fatal_signal_pending(current))
 457                                 return -ERESTARTNOHAND;
 458                         cond_resched();
 459                 }
 460         }
 461         return i;
 462 }
 463
 464 /*
 465  * 'copy_strings()' copies argument/environment strings from the old
 466  * processes's memory to the new process's stack.  The call to get_user_pages()
 467  * ensures the destination page is created and not swapped out.
 468  */
 469 static int copy_strings(int argc, struct user_arg_ptr argv,
 470                         struct linux_binprm *bprm)
 471 {
 472         struct page *kmapped_page = NULL;
 473         char *kaddr = NULL;
 474         unsigned long kpos = 0;
 475         int ret;
 476
 477         while (argc-- > 0) {
 478                 const char __user *str;
 479                 int len;
 480                 unsigned long pos;
 481
 482                 ret = -EFAULT;
 483                 str = get_user_arg_ptr(argv, argc);
 484                 if (IS_ERR(str))
 485                         goto out;
 486
 487                 len = strnlen_user(str, MAX_ARG_STRLEN);
 488                 if (!len)
 489                         goto out;
 490
 491                 ret = -E2BIG;
 492                 if (!valid_arg_len(bprm, len))
 493                         goto out;
 494
 495                 /* We're going to work our way backwords. */
 496                 pos = bprm->p;
 497                 str += len;
 498                 bprm->p -= len;
 499
 500                 while (len > 0) {
 501                         int offset, bytes_to_copy;
 502
 503                         if (fatal_signal_pending(current)) {
 504                                 ret = -ERESTARTNOHAND;
 505                                 goto out;
 506                         }
 507                         cond_resched();
 508
 509                         offset = pos % PAGE_SIZE;
 510                         if (offset == 0)
 511                                 offset = PAGE_SIZE;
 512
 513                         bytes_to_copy = offset;
 514                         if (bytes_to_copy > len)
 515                                 bytes_to_copy = len;
 516
 517                         offset -= bytes_to_copy;
 518                         pos -= bytes_to_copy;
 519                         str -= bytes_to_copy;
 520                         len -= bytes_to_copy;
 521
 522                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 523                                 struct page *page;
 524
 525                                 page = get_arg_page(bprm, pos, 1);
 526                                 if (!page) {
 527                                         ret = -E2BIG;
 528                                         goto out;
 529                                 }
 530
 531                                 if (kmapped_page) {
 532                                         flush_kernel_dcache_page(kmapped_page);
 533                                         kunmap(kmapped_page);
 534                                         put_arg_page(kmapped_page);
 535                                 }
 536                                 kmapped_page = page;
 537                                 kaddr = kmap(kmapped_page);
 538                                 kpos = pos & PAGE_MASK;
 539                                 flush_arg_page(bprm, kpos, kmapped_page);
 540                         }
 541                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 542                                 ret = -EFAULT;
 543                                 goto out;
 544                         }
 545                 }
 546         }
 547         ret = 0;
 548 out:
 549         if (kmapped_page) {
 550                 flush_kernel_dcache_page(kmapped_page);
 551                 kunmap(kmapped_page);
 552                 put_arg_page(kmapped_page);
 553         }
 554         return ret;
 555 }
 556
 557 /*
 558  * Like copy_strings, but get argv and its values from kernel memory.
 559  */
 560 int copy_strings_kernel(int argc, const char *const *__argv,
 561                         struct linux_binprm *bprm)
 562 {
 563         int r;
 564         mm_segment_t oldfs = get_fs();
 565         struct user_arg_ptr argv = {
 566                 .ptr.native = (const char __user *const  __user *)__argv,
 567         };
 568
 569         set_fs(KERNEL_DS);
 570         r = copy_strings(argc, argv, bprm);
 571         set_fs(oldfs);
 572
 573         return r;
 574 }
 575 EXPORT_SYMBOL(copy_strings_kernel);
 576
 577 #ifdef CONFIG_MMU
 578
 579 /*
 580  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 581  * the binfmt code determines where the new stack should reside, we shift it to
 582  * its final location.  The process proceeds as follows:
 583  *
 584  * 1) Use shift to calculate the new vma endpoints.
 585  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 586  *    arguments passed to subsequent functions are consistent.
 587  * 3) Move vma's page tables to the new range.
 588  * 4) Free up any cleared pgd range.
 589  * 5) Shrink the vma to cover only the new range.
 590  */
 591 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 592 {
 593         struct mm_struct *mm = vma->vm_mm;
 594         unsigned long old_start = vma->vm_start;
 595         unsigned long old_end = vma->vm_end;
 596         unsigned long length = old_end - old_start;
 597         unsigned long new_start = old_start - shift;
 598         unsigned long new_end = old_end - shift;
 599         struct mmu_gather tlb;
 600
 601         BUG_ON(new_start > new_end);
 602
 603         /*
 604          * ensure there are no vmas between where we want to go
 605          * and where we are
 606          */
 607         if (vma != find_vma(mm, new_start))
 608                 return -EFAULT;
 609
 610         /*
 611          * cover the whole range: [new_start, old_end)
 612          */
 613         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 614                 return -ENOMEM;
 615
 616         /*
 617          * move the page tables downwards, on failure we rely on
 618          * process cleanup to remove whatever mess we made.
 619          */
 620         if (length != move_page_tables(vma, old_start,
 621                                        vma, new_start, length))
 622                 return -ENOMEM;
 623
 624         lru_add_drain();
 625         tlb_gather_mmu(&tlb, mm, 0);
 626         if (new_end > old_start) {
 627                 /*
 628                  * when the old and new regions overlap clear from new_end.
 629                  */
 630                 free_pgd_range(&tlb, new_end, old_end, new_end,
 631                         vma->vm_next ? vma->vm_next->vm_start : 0);
 632         } else {
 633                 /*
 634                  * otherwise, clean from old_start; this is done to not touch
 635                  * the address space in [new_end, old_start) some architectures
 636                  * have constraints on va-space that make this illegal (IA64) -
 637                  * for the others its just a little faster.
 638                  */
 639                 free_pgd_range(&tlb, old_start, old_end, new_end,
 640                         vma->vm_next ? vma->vm_next->vm_start : 0);
 641         }
 642         tlb_finish_mmu(&tlb, new_end, old_end);
 643
 644         /*
 645          * Shrink the vma to just the new range.  Always succeeds.
 646          */
 647         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 648
 649         return 0;
 650 }
 651
 652 /*
 653  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 654  * the stack is optionally relocated, and some extra space is added.
 655  */
 656 int setup_arg_pages(struct linux_binprm *bprm,
 657                     unsigned long stack_top,
 658                     int executable_stack)
 659 {
 660         unsigned long ret;
 661         unsigned long stack_shift;
 662         struct mm_struct *mm = current->mm;
 663         struct vm_area_struct *vma = bprm->vma;
 664         struct vm_area_struct *prev = NULL;
 665         unsigned long vm_flags;
 666         unsigned long stack_base;
 667         unsigned long stack_size;
 668         unsigned long stack_expand;
 669         unsigned long rlim_stack;
 670
 671 #ifdef CONFIG_STACK_GROWSUP
 672         /* Limit stack size to 1GB */
 673         stack_base = rlimit_max(RLIMIT_STACK);
 674         if (stack_base > (1 << 30))
 675                 stack_base = 1 << 30;
 676
 677         /* Make sure we didn't let the argument array grow too large. */
 678         if (vma->vm_end - vma->vm_start > stack_base)
 679                 return -ENOMEM;
 680
 681         stack_base = PAGE_ALIGN(stack_top - stack_base);
 682
 683         stack_shift = vma->vm_start - stack_base;
 684         mm->arg_start = bprm->p - stack_shift;
 685         bprm->p = vma->vm_end - stack_shift;
 686 #else
 687         stack_top = arch_align_stack(stack_top);
 688         stack_top = PAGE_ALIGN(stack_top);
 689
 690         if (unlikely(stack_top < mmap_min_addr) ||
 691             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 692                 return -ENOMEM;
 693
 694         stack_shift = vma->vm_end - stack_top;
 695
 696         bprm->p -= stack_shift;
 697         mm->arg_start = bprm->p;
 698 #endif
 699
 700         if (bprm->loader)
 701                 bprm->loader -= stack_shift;
 702         bprm->exec -= stack_shift;
 703
 704         down_write(&mm->mmap_sem);
 705         vm_flags = VM_STACK_FLAGS;
 706
 707         /*
 708          * Adjust stack execute permissions; explicitly enable for
 709          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 710          * (arch default) otherwise.
 711          */
 712         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 713                 vm_flags |= VM_EXEC;
 714         else if (executable_stack == EXSTACK_DISABLE_X)
 715                 vm_flags &= ~VM_EXEC;
 716         vm_flags |= mm->def_flags;
 717         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 718
 719         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 720                         vm_flags);
 721         if (ret)
 722                 goto out_unlock;
 723         BUG_ON(prev != vma);
 724
 725         /* Move stack pages down in memory. */
 726         if (stack_shift) {
 727                 ret = shift_arg_pages(vma, stack_shift);
 728                 if (ret)
 729                         goto out_unlock;
 730         }
 731
 732         /* mprotect_fixup is overkill to remove the temporary stack flags */
 733         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 734
 735         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 736         stack_size = vma->vm_end - vma->vm_start;
 737         /*
 738          * Align this down to a page boundary as expand_stack
 739          * will align it up.
 740          */
 741         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 742 #ifdef CONFIG_STACK_GROWSUP
 743         if (stack_size + stack_expand > rlim_stack)
 744                 stack_base = vma->vm_start + rlim_stack;
 745         else
 746                 stack_base = vma->vm_end + stack_expand;
 747 #else
 748         if (stack_size + stack_expand > rlim_stack)
 749                 stack_base = vma->vm_end - rlim_stack;
 750         else
 751                 stack_base = vma->vm_start - stack_expand;
 752 #endif
 753         current->mm->start_stack = bprm->p;
 754         ret = expand_stack(vma, stack_base);
 755         if (ret)
 756                 ret = -EFAULT;
 757
 758 out_unlock:
 759         up_write(&mm->mmap_sem);
 760         return ret;
 761 }
 762 EXPORT_SYMBOL(setup_arg_pages);
 763
 764 #endif /* CONFIG_MMU */
 765
 766 struct file *open_exec(const char *name)
 767 {
 768         struct file *file;
 769         int err;
 770         static const struct open_flags open_exec_flags = {
 771                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 772                 .acc_mode = MAY_EXEC | MAY_OPEN,
 773                 .intent = LOOKUP_OPEN
 774         };
 775
 776         file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
 777         if (IS_ERR(file))
 778                 goto out;
 779
 780         err = -EACCES;
 781         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 782                 goto exit;
 783
 784         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 785                 goto exit;
 786
 787         fsnotify_open(file);
 788
 789         err = deny_write_access(file);
 790         if (err)
 791                 goto exit;
 792
 793 out:
 794         return file;
 795
 796 exit:
 797         fput(file);
 798         return ERR_PTR(err);
 799 }
 800 EXPORT_SYMBOL(open_exec);
 801
 802 int kernel_read(struct file *file, loff_t offset,
 803                 char *addr, unsigned long count)
 804 {
 805         mm_segment_t old_fs;
 806         loff_t pos = offset;
 807         int result;
 808
 809         old_fs = get_fs();
 810         set_fs(get_ds());
 811         /* The cast to a user pointer is valid due to the set_fs() */
 812         result = vfs_read(file, (void __user *)addr, count, &pos);
 813         set_fs(old_fs);
 814         return result;
 815 }
 816
 817 EXPORT_SYMBOL(kernel_read);
 818
 819 static int exec_mmap(struct mm_struct *mm)
 820 {
 821         struct task_struct *tsk;
 822         struct mm_struct * old_mm, *active_mm;
 823
 824         /* Notify parent that we're no longer interested in the old VM */
 825         tsk = current;
 826         old_mm = current->mm;
 827         sync_mm_rss(tsk, old_mm);
 828         mm_release(tsk, old_mm);
 829
 830         if (old_mm) {
 831                 /*
 832                  * Make sure that if there is a core dump in progress
 833                  * for the old mm, we get out and die instead of going
 834                  * through with the exec.  We must hold mmap_sem around
 835                  * checking core_state and changing tsk->mm.
 836                  */
 837                 down_read(&old_mm->mmap_sem);
 838                 if (unlikely(old_mm->core_state)) {
 839                         up_read(&old_mm->mmap_sem);
 840                         return -EINTR;
 841                 }
 842         }
 843         task_lock(tsk);
 844         active_mm = tsk->active_mm;
 845         tsk->mm = mm;
 846         tsk->active_mm = mm;
 847         activate_mm(active_mm, mm);
 848         task_unlock(tsk);
 849         arch_pick_mmap_layout(mm);
 850         if (old_mm) {
 851                 up_read(&old_mm->mmap_sem);
 852                 BUG_ON(active_mm != old_mm);
 853                 mm_update_next_owner(old_mm);
 854                 mmput(old_mm);
 855                 return 0;
 856         }
 857         mmdrop(active_mm);
 858         return 0;
 859 }
 860
 861 /*
 862  * This function makes sure the current process has its own signal table,
 863  * so that flush_signal_handlers can later reset the handlers without
 864  * disturbing other processes.  (Other processes might share the signal
 865  * table via the CLONE_SIGHAND option to clone().)
 866  */
 867 static int de_thread(struct task_struct *tsk)
 868 {
 869         struct signal_struct *sig = tsk->signal;
 870         struct sighand_struct *oldsighand = tsk->sighand;
 871         spinlock_t *lock = &oldsighand->siglock;
 872
 873         if (thread_group_empty(tsk))
 874                 goto no_thread_group;
 875
 876         /*
 877          * Kill all other threads in the thread group.
 878          */
 879         spin_lock_irq(lock);
 880         if (signal_group_exit(sig)) {
 881                 /*
 882                  * Another group action in progress, just
 883                  * return so that the signal is processed.
 884                  */
 885                 spin_unlock_irq(lock);
 886                 return -EAGAIN;
 887         }
 888
 889         sig->group_exit_task = tsk;
 890         sig->notify_count = zap_other_threads(tsk);
 891         if (!thread_group_leader(tsk))
 892                 sig->notify_count--;
 893
 894         while (sig->notify_count) {
 895                 __set_current_state(TASK_UNINTERRUPTIBLE);
 896                 spin_unlock_irq(lock);
 897                 schedule();
 898                 spin_lock_irq(lock);
 899         }
 900         spin_unlock_irq(lock);
 901
 902         /*
 903          * At this point all other threads have exited, all we have to
 904          * do is to wait for the thread group leader to become inactive,
 905          * and to assume its PID:
 906          */
 907         if (!thread_group_leader(tsk)) {
 908                 struct task_struct *leader = tsk->group_leader;
 909
 910                 sig->notify_count = -1; /* for exit_notify() */
 911                 for (;;) {
 912                         write_lock_irq(&tasklist_lock);
 913                         if (likely(leader->exit_state))
 914                                 break;
 915                         __set_current_state(TASK_UNINTERRUPTIBLE);
 916                         write_unlock_irq(&tasklist_lock);
 917                         schedule();
 918                 }
 919
 920                 /*
 921                  * The only record we have of the real-time age of a
 922                  * process, regardless of execs it's done, is start_time.
 923                  * All the past CPU time is accumulated in signal_struct
 924                  * from sister threads now dead.  But in this non-leader
 925                  * exec, nothing survives from the original leader thread,
 926                  * whose birth marks the true age of this process now.
 927                  * When we take on its identity by switching to its PID, we
 928                  * also take its birthdate (always earlier than our own).
 929                  */
 930                 tsk->start_time = leader->start_time;
 931
 932                 BUG_ON(!same_thread_group(leader, tsk));
 933                 BUG_ON(has_group_leader_pid(tsk));
 934                 /*
 935                  * An exec() starts a new thread group with the
 936                  * TGID of the previous thread group. Rehash the
 937                  * two threads with a switched PID, and release
 938                  * the former thread group leader:
 939                  */
 940
 941                 /* Become a process group leader with the old leader's pid.
 942                  * The old leader becomes a thread of the this thread group.
 943                  * Note: The old leader also uses this pid until release_task
 944                  *       is called.  Odd but simple and correct.
 945                  */
 946                 detach_pid(tsk, PIDTYPE_PID);
 947                 tsk->pid = leader->pid;
 948                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 949                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 950                 transfer_pid(leader, tsk, PIDTYPE_SID);
 951
 952                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 953                 list_replace_init(&leader->sibling, &tsk->sibling);
 954
 955                 tsk->group_leader = tsk;
 956                 leader->group_leader = tsk;
 957
 958                 tsk->exit_signal = SIGCHLD;
 959                 leader->exit_signal = -1;
 960
 961                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 962                 leader->exit_state = EXIT_DEAD;
 963
 964                 /*
 965                  * We are going to release_task()->ptrace_unlink() silently,
 966                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
 967                  * the tracer wont't block again waiting for this thread.
 968                  */
 969                 if (unlikely(leader->ptrace))
 970                         __wake_up_parent(leader, leader->parent);
 971                 write_unlock_irq(&tasklist_lock);
 972
 973                 release_task(leader);
 974         }
 975
 976         sig->group_exit_task = NULL;
 977         sig->notify_count = 0;
 978
 979 no_thread_group:
 980         /* we have changed execution domain */
 981         tsk->exit_signal = SIGCHLD;
 982
 983         if (current->mm)
 984                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
 985
 986         exit_itimers(sig);
 987         flush_itimer_signals();
 988
 989         if (atomic_read(&oldsighand->count) != 1) {
 990                 struct sighand_struct *newsighand;
 991                 /*
 992                  * This ->sighand is shared with the CLONE_SIGHAND
 993                  * but not CLONE_THREAD task, switch to the new one.
 994                  */
 995                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 996                 if (!newsighand)
 997                         return -ENOMEM;
 998
 999                 atomic_set(&newsighand->count, 1);
1000                 memcpy(newsighand->action, oldsighand->action,
1001                        sizeof(newsighand->action));
1002
1003                 write_lock_irq(&tasklist_lock);
1004                 spin_lock(&oldsighand->siglock);
1005                 rcu_assign_pointer(tsk->sighand, newsighand);
1006                 spin_unlock(&oldsighand->siglock);
1007                 write_unlock_irq(&tasklist_lock);
1008
1009                 __cleanup_sighand(oldsighand);
1010         }
1011
1012         BUG_ON(!thread_group_leader(tsk));
1013         return 0;
1014 }
1015
1016 /*
1017  * These functions flushes out all traces of the currently running executable
1018  * so that a new one can be started
1019  */
1020 static void flush_old_files(struct files_struct * files)
1021 {
1022         long j = -1;
1023         struct fdtable *fdt;
1024
1025         spin_lock(&files->file_lock);
1026         for (;;) {
1027                 unsigned long set, i;
1028
1029                 j++;
1030                 i = j * __NFDBITS;
1031                 fdt = files_fdtable(files);
1032                 if (i >= fdt->max_fds)
1033                         break;
1034                 set = fdt->close_on_exec->fds_bits[j];
1035                 if (!set)
1036                         continue;
1037                 fdt->close_on_exec->fds_bits[j] = 0;
1038                 spin_unlock(&files->file_lock);
1039                 for ( ; set ; i++,set >>= 1) {
1040                         if (set & 1) {
1041                                 sys_close(i);
1042                         }
1043                 }
1044                 spin_lock(&files->file_lock);
1045
1046         }
1047         spin_unlock(&files->file_lock);
1048 }
1049
1050 char *get_task_comm(char *buf, struct task_struct *tsk)
1051 {
1052         /* buf must be at least sizeof(tsk->comm) in size */
1053         task_lock(tsk);
1054         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1055         task_unlock(tsk);
1056         return buf;
1057 }
1058 EXPORT_SYMBOL_GPL(get_task_comm);
1059
1060 void set_task_comm(struct task_struct *tsk, char *buf)
1061 {
1062         task_lock(tsk);
1063
1064         /*
1065          * Threads may access current->comm without holding
1066          * the task lock, so write the string carefully.
1067          * Readers without a lock may see incomplete new
1068          * names but are safe from non-terminating string reads.
1069          */
1070         memset(tsk->comm, 0, TASK_COMM_LEN);
1071         wmb();
1072         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1073         task_unlock(tsk);
1074         perf_event_comm(tsk);
1075 }
1076
1077 int flush_old_exec(struct linux_binprm * bprm)
1078 {
1079         int retval;
1080
1081         /*
1082          * Make sure we have a private signal table and that
1083          * we are unassociated from the previous thread group.
1084          */
1085         retval = de_thread(current);
1086         if (retval)
1087                 goto out;
1088
1089         set_mm_exe_file(bprm->mm, bprm->file);
1090
1091         /*
1092          * Release all of the old mmap stuff
1093          */
1094         acct_arg_size(bprm, 0);
1095         retval = exec_mmap(bprm->mm);
1096         if (retval)
1097                 goto out;
1098
1099         bprm->mm = NULL;                /* We're using it now */
1100
1101         set_fs(USER_DS);
1102         current->flags &=
1103                 ~(PF_RANDOMIZE | PF_KTHREAD | PF_NOFREEZE | PF_FREEZER_NOSIG);
1104         flush_thread();
1105         current->personality &= ~bprm->per_clear;
1106
1107         return 0;
1108
1109 out:
1110         return retval;
1111 }
1112 EXPORT_SYMBOL(flush_old_exec);
1113
1114 void would_dump(struct linux_binprm *bprm, struct file *file)
1115 {
1116         if (inode_permission(file->f_path.dentry->d_inode, MAY_READ) < 0)
1117                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1118 }
1119 EXPORT_SYMBOL(would_dump);
1120
1121 void setup_new_exec(struct linux_binprm * bprm)
1122 {
1123         int i, ch;
1124         const char *name;
1125         char tcomm[sizeof(current->comm)];
1126
1127         arch_pick_mmap_layout(current->mm);
1128
1129         /* This is the point of no return */
1130         current->sas_ss_sp = current->sas_ss_size = 0;
1131
1132         if (current_euid() == current_uid() && current_egid() == current_gid())
1133                 set_dumpable(current->mm, 1);
1134         else
1135                 set_dumpable(current->mm, suid_dumpable);
1136
1137         name = bprm->filename;
1138
1139         /* Copies the binary name from after last slash */
1140         for (i=0; (ch = *(name++)) != '\0';) {
1141                 if (ch == '/')
1142                         i = 0; /* overwrite what we wrote */
1143                 else
1144                         if (i < (sizeof(tcomm) - 1))
1145                                 tcomm[i++] = ch;
1146         }
1147         tcomm[i] = '\0';
1148         set_task_comm(current, tcomm);
1149
1150         /* Set the new mm task size. We have to do that late because it may
1151          * depend on TIF_32BIT which is only updated in flush_thread() on
1152          * some architectures like powerpc
1153          */
1154         current->mm->task_size = TASK_SIZE;
1155
1156         /* install the new credentials */
1157         if (bprm->cred->uid != current_euid() ||
1158             bprm->cred->gid != current_egid()) {
1159                 current->pdeath_signal = 0;
1160         } else {
1161                 would_dump(bprm, bprm->file);
1162                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1163                         set_dumpable(current->mm, suid_dumpable);
1164         }
1165
1166         /* An exec changes our domain. We are no longer part of the thread
1167            group */
1168
1169         current->self_exec_id++;
1170
1171         flush_signal_handlers(current, 0);
1172         flush_old_files(current->files);
1173 }
1174 EXPORT_SYMBOL(setup_new_exec);
1175
1176 /*
1177  * Prepare credentials and lock ->cred_guard_mutex.
1178  * install_exec_creds() commits the new creds and drops the lock.
1179  * Or, if exec fails before, free_bprm() should release ->cred and
1180  * and unlock.
1181  */
1182 int prepare_bprm_creds(struct linux_binprm *bprm)
1183 {
1184         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1185                 return -ERESTARTNOINTR;
1186
1187         bprm->cred = prepare_exec_creds();
1188         if (likely(bprm->cred))
1189                 return 0;
1190
1191         mutex_unlock(&current->signal->cred_guard_mutex);
1192         return -ENOMEM;
1193 }
1194
1195 void free_bprm(struct linux_binprm *bprm)
1196 {
1197         free_arg_pages(bprm);
1198         if (bprm->cred) {
1199                 mutex_unlock(&current->signal->cred_guard_mutex);
1200                 abort_creds(bprm->cred);
1201         }
1202         /* If a binfmt changed the interp, free it. */
1203         if (bprm->interp != bprm->filename)
1204                 kfree(bprm->interp);
1205         kfree(bprm);
1206 }
1207
1208 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1209 {
1210         /* If a binfmt changed the interp, free it first. */
1211         if (bprm->interp != bprm->filename)
1212                 kfree(bprm->interp);
1213         bprm->interp = kstrdup(interp, GFP_KERNEL);
1214         if (!bprm->interp)
1215                 return -ENOMEM;
1216         return 0;
1217 }
1218 EXPORT_SYMBOL(bprm_change_interp);
1219
1220 /*
1221  * install the new credentials for this executable
1222  */
1223 void install_exec_creds(struct linux_binprm *bprm)
1224 {
1225         security_bprm_committing_creds(bprm);
1226
1227         commit_creds(bprm->cred);
1228         bprm->cred = NULL;
1229
1230         /*
1231          * Disable monitoring for regular users
1232          * when executing setuid binaries. Must
1233          * wait until new credentials are committed
1234          * by commit_creds() above
1235          */
1236         if (get_dumpable(current->mm) != SUID_DUMP_USER)
1237                 perf_event_exit_task(current);
1238         /*
1239          * cred_guard_mutex must be held at least to this point to prevent
1240          * ptrace_attach() from altering our determination of the task's
1241          * credentials; any time after this it may be unlocked.
1242          */
1243         security_bprm_committed_creds(bprm);
1244         mutex_unlock(&current->signal->cred_guard_mutex);
1245 }
1246 EXPORT_SYMBOL(install_exec_creds);
1247
1248 /*
1249  * determine how safe it is to execute the proposed program
1250  * - the caller must hold ->cred_guard_mutex to protect against
1251  *   PTRACE_ATTACH
1252  */
1253 int check_unsafe_exec(struct linux_binprm *bprm)
1254 {
1255         struct task_struct *p = current, *t;
1256         unsigned n_fs;
1257         int res = 0;
1258
1259         if (p->ptrace) {
1260                 if (p->ptrace & PT_PTRACE_CAP)
1261                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1262                 else
1263                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1264         }
1265
1266         n_fs = 1;
1267         spin_lock(&p->fs->lock);
1268         rcu_read_lock();
1269         for (t = next_thread(p); t != p; t = next_thread(t)) {
1270                 if (t->fs == p->fs)
1271                         n_fs++;
1272         }
1273         rcu_read_unlock();
1274
1275         if (p->fs->users > n_fs) {
1276                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1277         } else {
1278                 res = -EAGAIN;
1279                 if (!p->fs->in_exec) {
1280                         p->fs->in_exec = 1;
1281                         res = 1;
1282                 }
1283         }
1284         spin_unlock(&p->fs->lock);
1285
1286         return res;
1287 }
1288
1289 static void bprm_fill_uid(struct linux_binprm *bprm)
1290 {
1291         struct inode *inode;
1292         unsigned int mode;
1293         uid_t uid;
1294         gid_t gid;
1295
1296         /* clear any previous set[ug]id data from a previous binary */
1297         bprm->cred->euid = current_euid();
1298         bprm->cred->egid = current_egid();
1299
1300         if (bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)
1301                 return;
1302
1303         inode = bprm->file->f_path.dentry->d_inode;
1304         mode = ACCESS_ONCE(inode->i_mode);
1305         if (!(mode & (S_ISUID|S_ISGID)))
1306                 return;
1307
1308         /* Be careful if suid/sgid is set */
1309         mutex_lock(&inode->i_mutex);
1310
1311         /* reload atomically mode/uid/gid now that lock held */
1312         mode = inode->i_mode;
1313         uid = inode->i_uid;
1314         gid = inode->i_gid;
1315         mutex_unlock(&inode->i_mutex);
1316
1317         if (mode & S_ISUID) {
1318                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1319                 bprm->cred->euid = uid;
1320         }
1321
1322         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1323                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1324                 bprm->cred->egid = gid;
1325         }
1326 }
1327
1328 /*
1329  * Fill the binprm structure from the inode.
1330  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1331  *
1332  * This may be called multiple times for binary chains (scripts for example).
1333  */
1334 int prepare_binprm(struct linux_binprm *bprm)
1335 {
1336         int retval;
1337
1338         if (bprm->file->f_op == NULL)
1339                 return -EACCES;
1340
1341         bprm_fill_uid(bprm);
1342
1343         /* fill in binprm security blob */
1344         retval = security_bprm_set_creds(bprm);
1345         if (retval)
1346                 return retval;
1347         bprm->cred_prepared = 1;
1348
1349         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1350         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1351 }
1352
1353 EXPORT_SYMBOL(prepare_binprm);
1354
1355 /*
1356  * Arguments are '\0' separated strings found at the location bprm->p
1357  * points to; chop off the first by relocating brpm->p to right after
1358  * the first '\0' encountered.
1359  */
1360 int remove_arg_zero(struct linux_binprm *bprm)
1361 {
1362         int ret = 0;
1363         unsigned long offset;
1364         char *kaddr;
1365         struct page *page;
1366
1367         if (!bprm->argc)
1368                 return 0;
1369
1370         do {
1371                 offset = bprm->p & ~PAGE_MASK;
1372                 page = get_arg_page(bprm, bprm->p, 0);
1373                 if (!page) {
1374                         ret = -EFAULT;
1375                         goto out;
1376                 }
1377                 kaddr = kmap_atomic(page, KM_USER0);
1378
1379                 for (; offset < PAGE_SIZE && kaddr[offset];
1380                                 offset++, bprm->p++)
1381                         ;
1382
1383                 kunmap_atomic(kaddr, KM_USER0);
1384                 put_arg_page(page);
1385
1386                 if (offset == PAGE_SIZE)
1387                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1388         } while (offset == PAGE_SIZE);
1389
1390         bprm->p++;
1391         bprm->argc--;
1392         ret = 0;
1393
1394 out:
1395         return ret;
1396 }
1397 EXPORT_SYMBOL(remove_arg_zero);
1398
1399 /*
1400  * cycle the list of binary formats handler, until one recognizes the image
1401  */
1402 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1403 {
1404         unsigned int depth = bprm->recursion_depth;
1405         int try,retval;
1406         struct linux_binfmt *fmt;
1407         pid_t old_pid;
1408
1409         /* This allows 4 levels of binfmt rewrites before failing hard. */
1410         if (depth > 5)
1411                 return -ELOOP;
1412
1413         retval = security_bprm_check(bprm);
1414         if (retval)
1415                 return retval;
1416
1417         retval = audit_bprm(bprm);
1418         if (retval)
1419                 return retval;
1420
1421         /* Need to fetch pid before load_binary changes it */
1422         rcu_read_lock();
1423         old_pid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1424         rcu_read_unlock();
1425
1426         retval = -ENOENT;
1427         for (try=0; try<2; try++) {
1428                 read_lock(&binfmt_lock);
1429                 list_for_each_entry(fmt, &formats, lh) {
1430                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1431                         if (!fn)
1432                                 continue;
1433                         if (!try_module_get(fmt->module))
1434                                 continue;
1435                         read_unlock(&binfmt_lock);
1436                         bprm->recursion_depth = depth + 1;
1437                         retval = fn(bprm, regs);
1438                         bprm->recursion_depth = depth;
1439                         if (retval >= 0) {
1440                                 if (depth == 0)
1441                                         ptrace_event(PTRACE_EVENT_EXEC,
1442                                                         old_pid);
1443                                 put_binfmt(fmt);
1444                                 allow_write_access(bprm->file);
1445                                 if (bprm->file)
1446                                         fput(bprm->file);
1447                                 bprm->file = NULL;
1448                                 current->did_exec = 1;
1449                                 proc_exec_connector(current);
1450                                 return retval;
1451                         }
1452                         read_lock(&binfmt_lock);
1453                         put_binfmt(fmt);
1454                         if (retval != -ENOEXEC || bprm->mm == NULL)
1455                                 break;
1456                         if (!bprm->file) {
1457                                 read_unlock(&binfmt_lock);
1458                                 return retval;
1459                         }
1460                 }
1461                 read_unlock(&binfmt_lock);
1462 #ifdef CONFIG_MODULES
1463                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1464                         break;
1465                 } else {
1466 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1467                         if (printable(bprm->buf[0]) &&
1468                             printable(bprm->buf[1]) &&
1469                             printable(bprm->buf[2]) &&
1470                             printable(bprm->buf[3]))
1471                                 break; /* -ENOEXEC */
1472                         if (try)
1473                                 break; /* -ENOEXEC */
1474                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1475                 }
1476 #else
1477                 break;
1478 #endif
1479         }
1480         return retval;
1481 }
1482
1483 EXPORT_SYMBOL(search_binary_handler);
1484
1485 /*
1486  * sys_execve() executes a new program.
1487  */
1488 static int do_execve_common(const char *filename,
1489                                 struct user_arg_ptr argv,
1490                                 struct user_arg_ptr envp,
1491                                 struct pt_regs *regs)
1492 {
1493         struct linux_binprm *bprm;
1494         struct file *file;
1495         struct files_struct *displaced;
1496         bool clear_in_exec;
1497         int retval;
1498         const struct cred *cred = current_cred();
1499
1500         /*
1501          * We move the actual failure in case of RLIMIT_NPROC excess from
1502          * set*uid() to execve() because too many poorly written programs
1503          * don't check setuid() return code.  Here we additionally recheck
1504          * whether NPROC limit is still exceeded.
1505          */
1506         if ((current->flags & PF_NPROC_EXCEEDED) &&
1507             atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1508                 retval = -EAGAIN;
1509                 goto out_ret;
1510         }
1511
1512         /* We're below the limit (still or again), so we don't want to make
1513          * further execve() calls fail. */
1514         current->flags &= ~PF_NPROC_EXCEEDED;
1515
1516         retval = unshare_files(&displaced);
1517         if (retval)
1518                 goto out_ret;
1519
1520         retval = -ENOMEM;
1521         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1522         if (!bprm)
1523                 goto out_files;
1524
1525         retval = prepare_bprm_creds(bprm);
1526         if (retval)
1527                 goto out_free;
1528
1529         retval = check_unsafe_exec(bprm);
1530         if (retval < 0)
1531                 goto out_free;
1532         clear_in_exec = retval;
1533         current->in_execve = 1;
1534
1535         file = open_exec(filename);
1536         retval = PTR_ERR(file);
1537         if (IS_ERR(file))
1538                 goto out_unmark;
1539
1540         sched_exec();
1541
1542         bprm->file = file;
1543         bprm->filename = filename;
1544         bprm->interp = filename;
1545
1546         retval = bprm_mm_init(bprm);
1547         if (retval)
1548                 goto out_file;
1549
1550         bprm->argc = count(argv, MAX_ARG_STRINGS);
1551         if ((retval = bprm->argc) < 0)
1552                 goto out;
1553
1554         bprm->envc = count(envp, MAX_ARG_STRINGS);
1555         if ((retval = bprm->envc) < 0)
1556                 goto out;
1557
1558         retval = prepare_binprm(bprm);
1559         if (retval < 0)
1560                 goto out;
1561
1562         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1563         if (retval < 0)
1564                 goto out;
1565
1566         bprm->exec = bprm->p;
1567         retval = copy_strings(bprm->envc, envp, bprm);
1568         if (retval < 0)
1569                 goto out;
1570
1571         retval = copy_strings(bprm->argc, argv, bprm);
1572         if (retval < 0)
1573                 goto out;
1574
1575         retval = search_binary_handler(bprm,regs);
1576         if (retval < 0)
1577                 goto out;
1578
1579         /* execve succeeded */
1580         current->fs->in_exec = 0;
1581         current->in_execve = 0;
1582         acct_update_integrals(current);
1583         free_bprm(bprm);
1584         if (displaced)
1585                 put_files_struct(displaced);
1586         return retval;
1587
1588 out:
1589         if (bprm->mm) {
1590                 acct_arg_size(bprm, 0);
1591                 mmput(bprm->mm);
1592         }
1593
1594 out_file:
1595         if (bprm->file) {
1596                 allow_write_access(bprm->file);
1597                 fput(bprm->file);
1598         }
1599
1600 out_unmark:
1601         if (clear_in_exec)
1602                 current->fs->in_exec = 0;
1603         current->in_execve = 0;
1604
1605 out_free:
1606         free_bprm(bprm);
1607
1608 out_files:
1609         if (displaced)
1610                 reset_files_struct(displaced);
1611 out_ret:
1612         return retval;
1613 }
1614
1615 int do_execve(const char *filename,
1616         const char __user *const __user *__argv,
1617         const char __user *const __user *__envp,
1618         struct pt_regs *regs)
1619 {
1620         struct user_arg_ptr argv = { .ptr.native = __argv };
1621         struct user_arg_ptr envp = { .ptr.native = __envp };
1622         return do_execve_common(filename, argv, envp, regs);
1623 }
1624
1625 #ifdef CONFIG_COMPAT
1626 int compat_do_execve(char *filename,
1627         compat_uptr_t __user *__argv,
1628         compat_uptr_t __user *__envp,
1629         struct pt_regs *regs)
1630 {
1631         struct user_arg_ptr argv = {
1632                 .is_compat = true,
1633                 .ptr.compat = __argv,
1634         };
1635         struct user_arg_ptr envp = {
1636                 .is_compat = true,
1637                 .ptr.compat = __envp,
1638         };
1639         return do_execve_common(filename, argv, envp, regs);
1640 }
1641 #endif
1642
1643 void set_binfmt(struct linux_binfmt *new)
1644 {
1645         struct mm_struct *mm = current->mm;
1646
1647         if (mm->binfmt)
1648                 module_put(mm->binfmt->module);
1649
1650         mm->binfmt = new;
1651         if (new)
1652                 __module_get(new->module);
1653 }
1654
1655 EXPORT_SYMBOL(set_binfmt);
1656
1657 static int expand_corename(struct core_name *cn)
1658 {
1659         char *old_corename = cn->corename;
1660
1661         cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1662         cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1663
1664         if (!cn->corename) {
1665                 kfree(old_corename);
1666                 return -ENOMEM;
1667         }
1668
1669         return 0;
1670 }
1671
1672 static int cn_printf(struct core_name *cn, const char *fmt, ...)
1673 {
1674         char *cur;
1675         int need;
1676         int ret;
1677         va_list arg;
1678
1679         va_start(arg, fmt);
1680         need = vsnprintf(NULL, 0, fmt, arg);
1681         va_end(arg);
1682
1683         if (likely(need < cn->size - cn->used - 1))
1684                 goto out_printf;
1685
1686         ret = expand_corename(cn);
1687         if (ret)
1688                 goto expand_fail;
1689
1690 out_printf:
1691         cur = cn->corename + cn->used;
1692         va_start(arg, fmt);
1693         vsnprintf(cur, need + 1, fmt, arg);
1694         va_end(arg);
1695         cn->used += need;
1696         return 0;
1697
1698 expand_fail:
1699         return ret;
1700 }
1701
1702 static void cn_escape(char *str)
1703 {
1704         for (; *str; str++)
1705                 if (*str == '/')
1706                         *str = '!';
1707 }
1708
1709 static int cn_print_exe_file(struct core_name *cn)
1710 {
1711         struct file *exe_file;
1712         char *pathbuf, *path;
1713         int ret;
1714
1715         exe_file = get_mm_exe_file(current->mm);
1716         if (!exe_file) {
1717                 char *commstart = cn->corename + cn->used;
1718                 ret = cn_printf(cn, "%s (path unknown)", current->comm);
1719                 cn_escape(commstart);
1720                 return ret;
1721         }
1722
1723         pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
1724         if (!pathbuf) {
1725                 ret = -ENOMEM;
1726                 goto put_exe_file;
1727         }
1728
1729         path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
1730         if (IS_ERR(path)) {
1731                 ret = PTR_ERR(path);
1732                 goto free_buf;
1733         }
1734
1735         cn_escape(path);
1736
1737         ret = cn_printf(cn, "%s", path);
1738
1739 free_buf:
1740         kfree(pathbuf);
1741 put_exe_file:
1742         fput(exe_file);
1743         return ret;
1744 }
1745
1746 /* format_corename will inspect the pattern parameter, and output a
1747  * name into corename, which must have space for at least
1748  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1749  */
1750 static int format_corename(struct core_name *cn, long signr)
1751 {
1752         const struct cred *cred = current_cred();
1753         const char *pat_ptr = core_pattern;
1754         int ispipe = (*pat_ptr == '|');
1755         int pid_in_pattern = 0;
1756         int err = 0;
1757
1758         cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1759         cn->corename = kmalloc(cn->size, GFP_KERNEL);
1760         cn->used = 0;
1761
1762         if (!cn->corename)
1763                 return -ENOMEM;
1764
1765         /* Repeat as long as we have more pattern to process and more output
1766            space */
1767         while (*pat_ptr) {
1768                 if (*pat_ptr != '%') {
1769                         if (*pat_ptr == 0)
1770                                 goto out;
1771                         err = cn_printf(cn, "%c", *pat_ptr++);
1772                 } else {
1773                         switch (*++pat_ptr) {
1774                         /* single % at the end, drop that */
1775                         case 0:
1776                                 goto out;
1777                         /* Double percent, output one percent */
1778                         case '%':
1779                                 err = cn_printf(cn, "%c", '%');
1780                                 break;
1781                         /* pid */
1782                         case 'p':
1783                                 pid_in_pattern = 1;
1784                                 err = cn_printf(cn, "%d",
1785                                               task_tgid_vnr(current));
1786                                 break;
1787                         /* uid */
1788                         case 'u':
1789                                 err = cn_printf(cn, "%d", cred->uid);
1790                                 break;
1791                         /* gid */
1792                         case 'g':
1793                                 err = cn_printf(cn, "%d", cred->gid);
1794                                 break;
1795                         /* signal that caused the coredump */
1796                         case 's':
1797                                 err = cn_printf(cn, "%ld", signr);
1798                                 break;
1799                         /* UNIX time of coredump */
1800                         case 't': {
1801                                 struct timeval tv;
1802                                 do_gettimeofday(&tv);
1803                                 err = cn_printf(cn, "%lu", tv.tv_sec);
1804                                 break;
1805                         }
1806                         /* hostname */
1807                         case 'h': {
1808                                 char *namestart = cn->corename + cn->used;
1809                                 down_read(&uts_sem);
1810                                 err = cn_printf(cn, "%s",
1811                                               utsname()->nodename);
1812                                 up_read(&uts_sem);
1813                                 cn_escape(namestart);
1814                                 break;
1815                         }
1816                         /* executable */
1817                         case 'e': {
1818                                 char *commstart = cn->corename + cn->used;
1819                                 err = cn_printf(cn, "%s", current->comm);
1820                                 cn_escape(commstart);
1821                                 break;
1822                         }
1823                         case 'E':
1824                                 err = cn_print_exe_file(cn);
1825                                 break;
1826                         /* core limit size */
1827                         case 'c':
1828                                 err = cn_printf(cn, "%lu",
1829                                               rlimit(RLIMIT_CORE));
1830                                 break;
1831                         default:
1832                                 break;
1833                         }
1834                         ++pat_ptr;
1835                 }
1836
1837                 if (err)
1838                         return err;
1839         }
1840
1841         /* Backward compatibility with core_uses_pid:
1842          *
1843          * If core_pattern does not include a %p (as is the default)
1844          * and core_uses_pid is set, then .%pid will be appended to
1845          * the filename. Do not do this for piped commands. */
1846         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1847                 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1848                 if (err)
1849                         return err;
1850         }
1851 out:
1852         return ispipe;
1853 }
1854
1855 static int zap_process(struct task_struct *start, int exit_code)
1856 {
1857         struct task_struct *t;
1858         int nr = 0;
1859
1860         start->signal->flags = SIGNAL_GROUP_EXIT;
1861         start->signal->group_exit_code = exit_code;
1862         start->signal->group_stop_count = 0;
1863
1864         t = start;
1865         do {
1866                 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1867                 if (t != current && t->mm) {
1868                         sigaddset(&t->pending.signal, SIGKILL);
1869                         signal_wake_up(t, 1);
1870                         nr++;
1871                 }
1872         } while_each_thread(start, t);
1873
1874         return nr;
1875 }
1876
1877 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1878                                 struct core_state *core_state, int exit_code)
1879 {
1880         struct task_struct *g, *p;
1881         unsigned long flags;
1882         int nr = -EAGAIN;
1883
1884         spin_lock_irq(&tsk->sighand->siglock);
1885         if (!signal_group_exit(tsk->signal)) {
1886                 mm->core_state = core_state;
1887                 nr = zap_process(tsk, exit_code);
1888         }
1889         spin_unlock_irq(&tsk->sighand->siglock);
1890         if (unlikely(nr < 0))
1891                 return nr;
1892
1893         if (atomic_read(&mm->mm_users) == nr + 1)
1894                 goto done;
1895         /*
1896          * We should find and kill all tasks which use this mm, and we should
1897          * count them correctly into ->nr_threads. We don't take tasklist
1898          * lock, but this is safe wrt:
1899          *
1900          * fork:
1901          *      None of sub-threads can fork after zap_process(leader). All
1902          *      processes which were created before this point should be
1903          *      visible to zap_threads() because copy_process() adds the new
1904          *      process to the tail of init_task.tasks list, and lock/unlock
1905          *      of ->siglock provides a memory barrier.
1906          *
1907          * do_exit:
1908          *      The caller holds mm->mmap_sem. This means that the task which
1909          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1910          *      its ->mm.
1911          *
1912          * de_thread:
1913          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1914          *      we must see either old or new leader, this does not matter.
1915          *      However, it can change p->sighand, so lock_task_sighand(p)
1916          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1917          *      it can't fail.
1918          *
1919          *      Note also that "g" can be the old leader with ->mm == NULL
1920          *      and already unhashed and thus removed from ->thread_group.
1921          *      This is OK, __unhash_process()->list_del_rcu() does not
1922          *      clear the ->next pointer, we will find the new leader via
1923          *      next_thread().
1924          */
1925         rcu_read_lock();
1926         for_each_process(g) {
1927                 if (g == tsk->group_leader)
1928                         continue;
1929                 if (g->flags & PF_KTHREAD)
1930                         continue;
1931                 p = g;
1932                 do {
1933                         if (p->mm) {
1934                                 if (unlikely(p->mm == mm)) {
1935                                         lock_task_sighand(p, &flags);
1936                                         nr += zap_process(p, exit_code);
1937                                         unlock_task_sighand(p, &flags);
1938                                 }
1939                                 break;
1940                         }
1941                 } while_each_thread(g, p);
1942         }
1943         rcu_read_unlock();
1944 done:
1945         atomic_set(&core_state->nr_threads, nr);
1946         return nr;
1947 }
1948
1949 static int coredump_wait(int exit_code, struct core_state *core_state)
1950 {
1951         struct task_struct *tsk = current;
1952         struct mm_struct *mm = tsk->mm;
1953         struct completion *vfork_done;
1954         int core_waiters = -EBUSY;
1955
1956         init_completion(&core_state->startup);
1957         core_state->dumper.task = tsk;
1958         core_state->dumper.next = NULL;
1959
1960         down_write(&mm->mmap_sem);
1961         if (!mm->core_state)
1962                 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1963         up_write(&mm->mmap_sem);
1964
1965         if (unlikely(core_waiters < 0))
1966                 goto fail;
1967
1968         /*
1969          * Make sure nobody is waiting for us to release the VM,
1970          * otherwise we can deadlock when we wait on each other
1971          */
1972         vfork_done = tsk->vfork_done;
1973         if (vfork_done) {
1974                 tsk->vfork_done = NULL;
1975                 complete(vfork_done);
1976         }
1977
1978         if (core_waiters > 0) {
1979                 freezer_do_not_count();
1980                 wait_for_completion(&core_state->startup);
1981                 freezer_count();
1982         }
1983 fail:
1984         return core_waiters;
1985 }
1986
1987 static void coredump_finish(struct mm_struct *mm)
1988 {
1989         struct core_thread *curr, *next;
1990         struct task_struct *task;
1991
1992         next = mm->core_state->dumper.next;
1993         while ((curr = next) != NULL) {
1994                 next = curr->next;
1995                 task = curr->task;
1996                 /*
1997                  * see exit_mm(), curr->task must not see
1998                  * ->task == NULL before we read ->next.
1999                  */
2000                 smp_mb();
2001                 curr->task = NULL;
2002                 wake_up_process(task);
2003         }
2004
2005         mm->core_state = NULL;
2006 }
2007
2008 /*
2009  * set_dumpable converts traditional three-value dumpable to two flags and
2010  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
2011  * these bits are not changed atomically.  So get_dumpable can observe the
2012  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
2013  * return either old dumpable or new one by paying attention to the order of
2014  * modifying the bits.
2015  *
2016  * dumpable |   mm->flags (binary)
2017  * old  new | initial interim  final
2018  * ---------+-----------------------
2019  *  0    1  |   00      01      01
2020  *  0    2  |   00      10(*)   11
2021  *  1    0  |   01      00      00
2022  *  1    2  |   01      11      11
2023  *  2    0  |   11      10(*)   00
2024  *  2    1  |   11      11      01
2025  *
2026  * (*) get_dumpable regards interim value of 10 as 11.
2027  */
2028 void set_dumpable(struct mm_struct *mm, int value)
2029 {
2030         switch (value) {
2031         case 0:
2032                 clear_bit(MMF_DUMPABLE, &mm->flags);
2033                 smp_wmb();
2034                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
2035                 break;
2036         case 1:
2037                 set_bit(MMF_DUMPABLE, &mm->flags);
2038                 smp_wmb();
2039                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
2040                 break;
2041         case 2:
2042                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
2043                 smp_wmb();
2044                 set_bit(MMF_DUMPABLE, &mm->flags);
2045                 break;
2046         }
2047 }
2048
2049 static int __get_dumpable(unsigned long mm_flags)
2050 {
2051         int ret;
2052
2053         ret = mm_flags & MMF_DUMPABLE_MASK;
2054         return (ret >= 2) ? 2 : ret;
2055 }
2056
2057 /*
2058  * This returns the actual value of the suid_dumpable flag. For things
2059  * that are using this for checking for privilege transitions, it must
2060  * test against SUID_DUMP_USER rather than treating it as a boolean
2061  * value.
2062  */
2063 int get_dumpable(struct mm_struct *mm)
2064 {
2065         return __get_dumpable(mm->flags);
2066 }
2067
2068 static void wait_for_dump_helpers(struct file *file)
2069 {
2070         struct pipe_inode_info *pipe;
2071
2072         pipe = file->f_path.dentry->d_inode->i_pipe;
2073
2074         pipe_lock(pipe);
2075         pipe->readers++;
2076         pipe->writers--;
2077
2078         while ((pipe->readers > 1) && (!signal_pending(current))) {
2079                 wake_up_interruptible_sync(&pipe->wait);
2080                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
2081                 pipe_wait(pipe);
2082         }
2083
2084         pipe->readers--;
2085         pipe->writers++;
2086         pipe_unlock(pipe);
2087
2088 }
2089
2090
2091 /*
2092  * umh_pipe_setup
2093  * helper function to customize the process used
2094  * to collect the core in userspace.  Specifically
2095  * it sets up a pipe and installs it as fd 0 (stdin)
2096  * for the process.  Returns 0 on success, or
2097  * PTR_ERR on failure.
2098  * Note that it also sets the core limit to 1.  This
2099  * is a special value that we use to trap recursive
2100  * core dumps
2101  */
2102 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
2103 {
2104         struct file *rp, *wp;
2105         struct fdtable *fdt;
2106         struct coredump_params *cp = (struct coredump_params *)info->data;
2107         struct files_struct *cf = current->files;
2108
2109         wp = create_write_pipe(0);
2110         if (IS_ERR(wp))
2111                 return PTR_ERR(wp);
2112
2113         rp = create_read_pipe(wp, 0);
2114         if (IS_ERR(rp)) {
2115                 free_write_pipe(wp);
2116                 return PTR_ERR(rp);
2117         }
2118
2119         cp->file = wp;
2120
2121         sys_close(0);
2122         fd_install(0, rp);
2123         spin_lock(&cf->file_lock);
2124         fdt = files_fdtable(cf);
2125         FD_SET(0, fdt->open_fds);
2126         FD_CLR(0, fdt->close_on_exec);
2127         spin_unlock(&cf->file_lock);
2128
2129         /* and disallow core files too */
2130         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
2131
2132         return 0;
2133 }
2134
2135 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
2136 {
2137         struct core_state core_state;
2138         struct core_name cn;
2139         struct mm_struct *mm = current->mm;
2140         struct linux_binfmt * binfmt;
2141         const struct cred *old_cred;
2142         struct cred *cred;
2143         int retval = 0;
2144         int ispipe;
2145         /* require nonrelative corefile path and be extra careful */
2146         bool need_suid_safe = false;
2147         static atomic_t core_dump_count = ATOMIC_INIT(0);
2148         struct coredump_params cprm = {
2149                 .signr = signr,
2150                 .regs = regs,
2151                 .limit = rlimit(RLIMIT_CORE),
2152                 /*
2153                  * We must use the same mm->flags while dumping core to avoid
2154                  * inconsistency of bit flags, since this flag is not protected
2155                  * by any locks.
2156                  */
2157                 .mm_flags = mm->flags,
2158         };
2159
2160         audit_core_dumps(signr);
2161
2162         binfmt = mm->binfmt;
2163         if (!binfmt || !binfmt->core_dump)
2164                 goto fail;
2165         if (!__get_dumpable(cprm.mm_flags))
2166                 goto fail;
2167
2168         cred = prepare_creds();
2169         if (!cred)
2170                 goto fail;
2171         /*
2172          * We cannot trust fsuid as being the "true" uid of the process
2173          * nor do we know its entire history. We only know it was tainted
2174          * so we dump it as root in mode 2, and only into a controlled
2175          * environment (pipe handler or fully qualified path).
2176          */
2177         if (__get_dumpable(cprm.mm_flags) == 2) {
2178                 /* Setuid core dump mode */
2179                 cred->fsuid = 0;        /* Dump root private */
2180                 need_suid_safe = true;
2181         }
2182
2183         retval = coredump_wait(exit_code, &core_state);
2184         if (retval < 0)
2185                 goto fail_creds;
2186
2187         old_cred = override_creds(cred);
2188
2189         /*
2190          * Clear any false indication of pending signals that might
2191          * be seen by the filesystem code called to write the core file.
2192          */
2193         clear_thread_flag(TIF_SIGPENDING);
2194
2195         ispipe = format_corename(&cn, signr);
2196
2197         if (ispipe) {
2198                 int dump_count;
2199                 char **helper_argv;
2200
2201                 if (ispipe < 0) {
2202                         printk(KERN_WARNING "format_corename failed\n");
2203                         printk(KERN_WARNING "Aborting core\n");
2204                         goto fail_corename;
2205                 }
2206
2207                 if (cprm.limit == 1) {
2208                         /*
2209                          * Normally core limits are irrelevant to pipes, since
2210                          * we're not writing to the file system, but we use
2211                          * cprm.limit of 1 here as a speacial value. Any
2212                          * non-1 limit gets set to RLIM_INFINITY below, but
2213                          * a limit of 0 skips the dump.  This is a consistent
2214                          * way to catch recursive crashes.  We can still crash
2215                          * if the core_pattern binary sets RLIM_CORE =  !1
2216                          * but it runs as root, and can do lots of stupid things
2217                          * Note that we use task_tgid_vnr here to grab the pid
2218                          * of the process group leader.  That way we get the
2219                          * right pid if a thread in a multi-threaded
2220                          * core_pattern process dies.
2221                          */
2222                         printk(KERN_WARNING
2223                                 "Process %d(%s) has RLIMIT_CORE set to 1\n",
2224                                 task_tgid_vnr(current), current->comm);
2225                         printk(KERN_WARNING "Aborting core\n");
2226                         goto fail_unlock;
2227                 }
2228                 cprm.limit = RLIM_INFINITY;
2229
2230                 dump_count = atomic_inc_return(&core_dump_count);
2231                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2232                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2233                                task_tgid_vnr(current), current->comm);
2234                         printk(KERN_WARNING "Skipping core dump\n");
2235                         goto fail_dropcount;
2236                 }
2237
2238                 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2239                 if (!helper_argv) {
2240                         printk(KERN_WARNING "%s failed to allocate memory\n",
2241                                __func__);
2242                         goto fail_dropcount;
2243                 }
2244
2245                 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2246                                         NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2247                                         NULL, &cprm);
2248                 argv_free(helper_argv);
2249                 if (retval) {
2250                         printk(KERN_INFO "Core dump to %s pipe failed\n",
2251                                cn.corename);
2252                         goto close_fail;
2253                 }
2254         } else {
2255                 struct inode *inode;
2256                 int open_flags = O_CREAT | O_RDWR | O_NOFOLLOW |
2257                                  O_LARGEFILE | O_EXCL;
2258
2259                 if (cprm.limit < binfmt->min_coredump)
2260                         goto fail_unlock;
2261
2262                 if (need_suid_safe && cn.corename[0] != '/') {
2263                         printk(KERN_WARNING "Pid %d(%s) can only dump core "\
2264                                 "to fully qualified path!\n",
2265                                 task_tgid_vnr(current), current->comm);
2266                         printk(KERN_WARNING "Skipping core dump\n");
2267                         goto fail_unlock;
2268                 }
2269
2270                 /*
2271                  * Unlink the file if it exists unless this is a SUID
2272                  * binary - in that case, we're running around with root
2273                  * privs and don't want to unlink another user's coredump.
2274                  */
2275                 if (!need_suid_safe) {
2276                         mm_segment_t old_fs;
2277
2278                         old_fs = get_fs();
2279                         set_fs(KERNEL_DS);
2280                         /*
2281                          * If it doesn't exist, that's fine. If there's some
2282                          * other problem, we'll catch it at the filp_open().
2283                          */
2284                         (void) sys_unlink((const char __user *)cn.corename);
2285                         set_fs(old_fs);
2286                 }
2287
2288                 /*
2289                  * There is a race between unlinking and creating the
2290                  * file, but if that causes an EEXIST here, that's
2291                  * fine - another process raced with us while creating
2292                  * the corefile, and the other process won. To userspace,
2293                  * what matters is that at least one of the two processes
2294                  * writes its coredump successfully, not which one.
2295                  */
2296                 if (need_suid_safe) {
2297                         /*
2298                          * Using user namespaces, normal user tasks can change
2299                          * their current->fs->root to point to arbitrary
2300                          * directories. Since the intention of the "only dump
2301                          * with a fully qualified path" rule is to control where
2302                          * coredumps may be placed using root privileges,
2303                          * current->fs->root must not be used. Instead, use the
2304                          * root directory of init_task.
2305                          */
2306                         struct path root;
2307
2308                         task_lock(&init_task);
2309                         get_fs_root(init_task.fs, &root);
2310                         task_unlock(&init_task);
2311                         cprm.file = file_open_root(root.dentry, root.mnt,
2312                                 cn.corename, open_flags, 0600);
2313                         path_put(&root);
2314                 } else {
2315                         cprm.file = filp_open(cn.corename, open_flags, 0600);
2316                 }
2317                 if (IS_ERR(cprm.file))
2318                         goto fail_unlock;
2319
2320                 inode = cprm.file->f_path.dentry->d_inode;
2321                 if (inode->i_nlink > 1)
2322                         goto close_fail;
2323                 if (d_unhashed(cprm.file->f_path.dentry))
2324                         goto close_fail;
2325                 /*
2326                  * AK: actually i see no reason to not allow this for named
2327                  * pipes etc, but keep the previous behaviour for now.
2328                  */
2329                 if (!S_ISREG(inode->i_mode))
2330                         goto close_fail;
2331                 /*
2332                  * Dont allow local users get cute and trick others to coredump
2333                  * into their pre-created files.
2334                  */
2335                 if (inode->i_uid != current_fsuid())
2336                         goto close_fail;
2337                 if (!cprm.file->f_op || !cprm.file->f_op->write)
2338                         goto close_fail;
2339                 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2340                         goto close_fail;
2341         }
2342
2343         retval = binfmt->core_dump(&cprm);
2344         if (retval)
2345                 current->signal->group_exit_code |= 0x80;
2346
2347         if (ispipe && core_pipe_limit)
2348                 wait_for_dump_helpers(cprm.file);
2349 close_fail:
2350         if (cprm.file)
2351                 filp_close(cprm.file, NULL);
2352 fail_dropcount:
2353         if (ispipe)
2354                 atomic_dec(&core_dump_count);
2355 fail_unlock:
2356         kfree(cn.corename);
2357 fail_corename:
2358         coredump_finish(mm);
2359         revert_creds(old_cred);
2360 fail_creds:
2361         put_cred(cred);
2362 fail:
2363         return;
2364 }
2365
2366 /*
2367  * Core dumping helper functions.  These are the only things you should
2368  * do on a core-file: use only these functions to write out all the
2369  * necessary info.
2370  */
2371 int dump_write(struct file *file, const void *addr, int nr)
2372 {
2373         return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2374 }
2375 EXPORT_SYMBOL(dump_write);
2376
2377 int dump_seek(struct file *file, loff_t off)
2378 {
2379         int ret = 1;
2380
2381         if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2382                 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2383                         return 0;
2384         } else {
2385                 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2386
2387                 if (!buf)
2388                         return 0;
2389                 while (off > 0) {
2390                         unsigned long n = off;
2391
2392                         if (n > PAGE_SIZE)
2393                                 n = PAGE_SIZE;
2394                         if (!dump_write(file, buf, n)) {
2395                                 ret = 0;
2396                                 break;
2397                         }
2398                         off -= n;
2399                 }
2400                 free_page((unsigned long)buf);
2401         }
2402         return ret;
2403 }
2404 EXPORT_SYMBOL(dump_seek);